An AI agent usually borrows your identity. It sends from your account, works inside your browser and appears as another thread under your name. Manus Cue proposes a different model: give the agent an identity of its own.
In Cue, an agent can have its own email address, phone number, wallet and computer. Put several agents in a group chat and they can divide work among themselves. You set direction, watch progress and make the final decision.
That is one of the boldest product designs in the current personal-agent wave. It also concentrates the hardest questions—authority, money, impersonation, memory and accountability—inside one friendly interface.
What Manus Cue Is
Manus announced Cue as part of Manus 2.0 on September 28, 2026. It is a standalone app for desktop and mobile, built on the same infrastructure as Manus but organized around persistent personal agents rather than a general project workspace.
The product is in early access and requires an invite code. Manus initially described it as free for invited users, with web, desktop and mobile access and iOS availability following App Store review. Those rollout details can change quickly, so availability should always be checked in the live product.
Cue shares an account and data relationship with Manus. The Manus account-deletion guidance says a separate Cue-only account deletion is not currently supported. That detail matters: “standalone app” describes the interface and product experience, not complete separation of identity or stored data.
Every Agent Gets an Operational Identity
Cue’s defining claim is that each agent has four resources:
- an email address;
- a phone number;
- a wallet;
- a computer.
Together, these turn a character into an operator. Email and phone let the agent communicate through channels used by people and businesses. The computer gives it a place to browse and complete work. The wallet allows payment within a budget the user sets.
This is more consequential than giving an assistant a name or avatar. A separate address makes it possible to include the agent in a thread without pretending the message came directly from you. A separate number can take calls and leave a summary. A bounded wallet can make the spending limit explicit instead of handing over a general-purpose payment account.
The design could make accountability clearer, but identity alone does not guarantee it. Recipients still need to understand that they are dealing with an agent. Users need records of what was sent, promised and purchased. A budget prevents unlimited spending; it does not determine whether a purchase was appropriate.
From One Assistant to a Personal Team
Cue allows multiple agents to share a group chat and a goal. Manus gives the example of a launch project in which one agent researches venues, another creates a shortlist and a third drafts the presentation.
The important feature is the handoff. Each agent is not merely producing an independent answer; work moves from one role to the next. The user sets direction and makes the final call instead of manually copying every artifact.
This mirrors Grok Bot, which also presents agents as a roster of teammates. Cue differentiates itself by emphasizing external identity: phone, email and wallet are first-class product objects, not merely connected tools. Grok Bot emphasizes persistent cloud workspaces, learned routines and organizational collaboration.
In both products, team structure can hide errors as easily as it hides complexity. A researcher who starts from the wrong premise can send a polished mistake downstream. Good delegation therefore requires an evidence trail, unresolved-question list and approval boundary at every consequential handoff.
Calls, Payments and Everyday Services
Manus says a Cue agent can take calls and leave a summary in the app. It also describes everyday service interactions: scan a restaurant QR code, then let the agent order or hold a place in line.
These examples reveal Cue’s ambition. The agent is not confined to knowledge work; it is meant to interact with the service layer of daily life. That puts it closer to a personal representative than a productivity chatbot.
But the last mile is where product claims become difficult. Restaurants, phone systems, regional payment rules, merchant disputes and identity verification do not behave consistently. A useful review must separate the existence of a feature from its coverage. “Can place an order” does not mean every merchant, country or edge case works.
Payments deserve the strictest treatment. Before funding an agent wallet, users should verify transaction limits, confirmation rules, receipts, refunds, dispute handling and what happens when the agent misunderstands a request. The launch post establishes the direction; it does not answer every consumer-protection question.
Scheduling and Proactive Work
Cue can create scheduled automations from natural-language instructions. That allows an agent to do more than wait in a chat: it can prepare a recurring brief, check a condition or begin a task at a chosen time.
Manus 2.0 also introduced broader Automations that can respond to events in connected services, including email, calendar, Slack, Notion and changes in advertising performance. Cue sits on the personal-agent side of that infrastructure, while Manus Studio and Cloud Computer serve broader project work.
Proactivity is useful only when the user can answer three questions: what triggered this work, what data did it use, and what may it do without approval? A schedule should not become a standing blank check. Read-only monitoring, drafting and external action should remain separate permissions.
What Is Available and What Still Needs Evidence
As of October 6, 2026, the official picture is clear but narrow:
- Cue is a real standalone app in early access.
- Access requires an invite code.
- Agents can be organized as individuals or teams.
- Manus describes per-agent email, phone, wallet and computer identities.
- Scheduled automation is supported through natural-language setup.
- Manus and Cue share the same account and data relationship.
What remains unclear from public material is just as important: supported countries and phone regions, wallet providers and protections, detailed per-action approval controls, audit and export depth, long-term memory controls, enterprise administration and independent task-success data.
Those unknowns do not make the product imaginary. They make it early. The article should be updated as documentation replaces launch language and as real users publish reproducible tests.
Manus Cue vs. Grok Bot, OpenAI Dots and Today
| Decision | Manus Cue | [Grok Bot](/blog/what-is-grok-bot) | [OpenAI Dots](/blog/what-is-openai-dots) | [Today](/blog/what-is-today) |
|---|---|---|---|---|
| Primary object | Personal team of agents with identities | Roster of persistent digital coworkers | One always-on dot | One whole-life personal agent |
| Distinctive capability | Email, phone, wallet and computer per agent | Cloud computer, multi-Bot coordination and learned routines | ChatGPT continuity, plugins and cloud execution | Inspectable living memory and whole-day continuity |
| Collaboration | Agents hand work through group chats | Bots coordinate and transfer ownership | A dot delegates background tasks | User remains the center of one continuous context |
| Access posture | Invite-only early access | Beta through eligible paid Grok and Cursor plans | Eligible ChatGPT plans and regions | Available across Mac, iPhone and Android |
| Main trust question | What can an agent identity promise or purchase? | How are persistent sessions and shared work governed? | How inspectable and correctable is dot memory? | How should proactive help stay useful without becoming invasive? |
Cue is the clearest choice for someone specifically interested in agents that can represent themselves in communication and transactions. Grok Bot is stronger conceptually for repeat work performed by a digital team. Dots fit users already centered on ChatGPT. Today is for people who want one agent to understand a life spanning work, health and family without turning every domain into a separate worker.
How to Trial Cue Safely
If you receive an invite, begin with one agent and one reversible responsibility.
1. Use a new, low-risk communication channel rather than forwarding an entire personal inbox. 2. Keep the wallet empty until you have inspected the approval and receipt flow. 3. Ask the agent to draft rather than send during the first week. 4. Review every handoff when several agents collaborate. 5. Test deletion, export and disconnection before valuable history accumulates.
The purpose of an early trial is not to prove that the agent can do something once. It is to learn whether the system behaves predictably when context is incomplete, a service changes, a recipient responds unexpectedly or a task cannot be finished.
Identity Makes the Agent More Useful—and More Accountable
Cue identifies an important missing layer in personal AI. If an agent is expected to participate in the world, it needs more than intelligence. It needs channels, resources, boundaries and a legible identity.
Giving each agent an email, phone, wallet and computer makes the promise tangible. It also makes mistakes tangible. A wrong answer in chat is inconvenient; a wrong promise, call or payment can affect another person.
That is why Cue’s long-term quality will be judged less by how many identities users can create than by how clearly authority travels with them. The winning personal-agent product will not be the one that acts most often. It will be the one that lets users understand who acted, why, with what permission—and what can still be undone.